Integracion Richardson y Roomberg

Numerical Methods
Lecture 5.
Numerical integration
dr hab. inż. Katarzyna Zakrzewska, prof. AGH
Numerical Methods lecture 5 1

Outline
• Trapezoidal rule
• Multi-segment trapezoidal rule
• Richardson extrapolation
• Romberg's method
• Simpson's rule
• Gaussian method

Numerical integration - idea
b
I = ∫ f ( x )dx Y
f(x)
a
The integral can

be approximated
b
by: ∫ f ( x)dx
a
n
S = ∑ f (ci )Δxi a b X
i =1
xi ≤ ci ≤ xi +1

Newton–Cotes methods
Newton – Cotes integration belongs to a class of

methods with fixed nodes: function f(x) is interpolated
by a polynomial (e.g. Lagrange polynomial)
f ( x ) ≈ fn( x )
where: n −1
f n ( x ) = a0 + a1 x + ... + an −1 x + an x n
Then, the integral of f(x) can be approximated as an

integral of the interpolated function fn(x)
b b
I = ∫ f ( x) dx ≈ ∫ f n ( x) dx
a a

Trapezoidal rule
The trapezoidal rule f1 ( x) = a0 + a1 x

assumes: n = 1, thus:
b b b
I = ∫ f ( x) dx ≈ ∫ f1 ( x) dx = ∫ (a0 + a1 x)dx
a a a
b2 − a 2
= a0 (b − a ) + a1
2
But what is a0 and a1 ?
Now if one chooses, (a,f(a)) i (b,f(b)) as the two points to
approximate f(x) by a straight line from a to b. It follows that:
f (a)b − f (b)a
f (a ) = f1 (a ) = a0 + a1a a0 =
b−a
f (b) = f1 (b) = a0 + a1b f (b) − f (a)
a1 =
b−a
Trapezoidal rule
b
⎡ f ( a ) + f (b ) ⎤ b
∫ f ( x ) dx = (b − a ) ⎢
a ⎣ 2 ⎥⎦ ∫ f ( x)dx ≈ area of trapezoid
a
Y
f(x)
f(b) fI(x)
f(a)
a b X

Trapezoidal rule
Example 1:
Velocity v(t) of a rocket from t1=8 s to t2=30 s is given by:
⎡ 140000 ⎤
v(t ) = 2000 ln ⎢ ⎥ − 9.8t
⎣140000 − 2100t ⎦
a) Use the single segment trapezoidal rule to find the distance

covered by the rocket from t1=8 s to t2=30 s
b) Find the true relative error.

Trapezoidal rule
a) ⎡ f (a ) + f (b) ⎤
I ≈ (b − a ) ⎢ ⎥ a =8 s b = 30 s
⎣ 2 ⎦
⎡ 140000 ⎤
f (t ) = 2000 ln ⎢ ⎥ − 9.8t
⎣140000 − 2100t ⎦
⎡ 140000 ⎤
f (8) = 2000 ln ⎢ ⎥ − 9.8(8) = 177.27 m / s
⎣140000 − 2100(8) ⎦
⎡ 140000 ⎤
f (30) = 2000 ln ⎢ ⎥ − 9.8(30) = 901.67 m / s
⎣140000 − 2100(30) ⎦
⎡177.27 + 901.67 ⎤ = 11868 m

I = (30 − 8) ⎢ ⎥⎦
⎣ 2

Trapezoidal rule
b) The true value
30⎛ ⎞
⎡ 140000 ⎤
Δx = ∫ ⎜⎜ 2000 ln ⎢ ⎥ − 9.8t ⎟⎟dt = 11061 m
8⎝ ⎣140000 − 2100t ⎦ ⎠
The relative error:
11061 − 11868
∈t = × 100 = 7 .2959 %
11061

Multi-segment trapezoidal rule
The true error using a

Y
single segment trapezoidal f(x)
rule was large. We can
divide the interval from a to
b into smaller n segments
of equal length – h and
apply the trapezoidal rule
over each segment:
b−a
h=
n a b−a a+2
b−a
a+3
b−a b X
a+ 4
4
for n=4 4
b a+h a+2h a + 3h a+4h

I = ∫ f ( x)dx = ∫ f ( x)dx + ∫ f ( x)dx + ∫ f ( x)dx + ∫ f ( x)dx
a a a+h a+2h a + 3h

b a+h a + 2h a + ( n −1 ) h b
∫ f ( x )dx = ∫ f ( x )dx + ∫ f ( x )dx + ... + ∫ f ( x )dx + ∫ f ( x )dx
a a a+h a + ( n − 2 )h a + ( n −1 )h
⎡ f (a ) + f (a + h) ⎤ ⎡ f (a + h) + f (a + 2h) ⎤
= h⎢ ⎥ + h⎢ ⎥ + ...
⎣ 2 ⎦ ⎣ 2 ⎦
⎡ f ( a + ( n − 1) h ) + f (b ) ⎤
... + [b − ( a + ( n − 1) h ]⎢ ⎥
⎣ 2 ⎦
b
b−a⎡ ⎧ n −1
⎫ ⎤
∫ f ( x )dx = ⎢ f ( a ) + 2⎨ ∑ f ( a + ih )⎬ + f ( b )⎥
a 2n ⎣ ⎩ i =1 ⎭ ⎦

Example 2:
⎡ 140000 ⎤
v(t ) = 2000 ln ⎢ ⎥ − 9.8t
⎣140000 − 2100t ⎦
a) Use the complex segment trapezoidal rule to find the distance

covered from t1=8 s to t2=30 s for n = 2
b) Find the true relative error.

b−a ⎡ ⎧n −1 ⎫ ⎤
a) I= ⎢ f ( a ) + 2 ⎨∑ f ( a + ih ) ⎬ + f (b ) ⎥
2n ⎣ ⎩ i =1 ⎭ ⎦
b − a 30 − 8
n=2 a=8s b = 30 s h= = = 11s
n 2
30 − 8 ⎡ ⎧2−1 ⎫ ⎤
I= ⎢ f (8) + 2⎨ ∑ f (a + ih)⎬ + f (30)⎥
2( 2) ⎣ ⎩ i =1 ⎭ ⎦
22 22
= [ f (8) + 2 f (19) + f (30)] = [177.27 + 2( 484.75 ) + 901.67]
4 4
= 11266 m

b) true value:
30
⎛ ⎡ 140000 ⎤ ⎞
x = ∫ ⎜ 2000ln⎢ ⎥ − 9.8t ⎟dt = 11061 m
8⎝ ⎣140000 − 2100t ⎦ ⎠
The relative error:
11061 − 11266
∈t = × 100 = 1.8534%
11061

n ∆x Et ∈t % ∈a %
1 11868 -807 7.296 ---
2 11266 -205 1.853 5.343
3 11153 -91.4 0.8265 1.019
4 11113 -51.5 0.4655 0.3594
5 11094 -33.0 0.2981 0.1669
6 11084 -22.9 0.2070 0.09082
7 11078 -16.8 0.1521 0.05482
8 11074 -12.9 0.1165 0.03560

Estimation of error
The relative error for a simple trapezoidal rule is
( b − a )3
Et = f " ( ζ ), a < ζ < b
12
The relative error in the multi-segment trapezoidal rule is a sum of errors
for each segment. The relative error within the first segment is given by:
[( a + h ) − a ]3
E1 = f " ( ζ 1 ), a < ζ 1 < a + h
12
h3
= f " ( ζ1 )
12

Estimation of error
By analogy:
Ei =
[(a + ih) − (a + (i − 1)h)]
3
f " (ζ i ), a + (i − 1)h < ζ i < a + ih
12
h3
= f " ( ζi )
12
for n-th segment :
[b − {a + ( n − 1 )h}]3
En = f " ( ζ n ), a + ( n − 1 )h < ζ n < b
12
h3
= f" (ζn )
12

Estimation of error
The total error in the complex trapezoidal rule is a sum of the errors for a single
segment:
n
3 ∑ f " ( ζi )
n h3 n (b − a ) i =1
Et = ∑ Ei = ∑ f " ( ζi ) =
i =1
12 i =1 12n 2 n
n
∑ f" (ζi )
Formula: i =1
n
yields an approximate average value of the second derivative in the range of
a< x<b
1
Et ∝ α 2
n
Estimation of error
Table below presents the results for the integral

30
⎛ ⎡ 140000 ⎤ ⎞
∫ ⎜⎝ 2000 ln ⎢⎣140000 − 2100t ⎥⎦ − 9. 8t ⎟dt
⎠
8
as a function of the number of segments n. When n twice increases, the
absolute error Et decreases four times!
n Value Et ∈t % ∈a %
2 11266 -205 1.854 5.343
4 11113 -51.5 0.4655 0.3594
8 11074 -12.9 0.1165 0.03560
16 11065 -3.22 0.02913 0.00401

Richardson’s extrapolation and Romberg’s method
of integration
Richardson’s extrapolation and Romberg’s method

of integration constitute an extension of the
trapezoidal method and give better approximation
of the integral by reducing the true error.

Richardson extrapolation
The true error obtained when using the multi-segment

trapezoidal rule with n segments to approximate an integral is
given by: C
Et ≅
n2
where: C is an approximate constant of proportionality
Since:
Et = TV − I n
approximate value using

true value
n-segments

It can be shown that: C
≅ TV − I 2 n
(2 n ) 2
If the number of segments is doubled from n to 2n:
C
≅ TV − I n
(n ) 2
C
≅ TV − I 2 n
(2 n ) 2
I 2n − I n
We get: TV ≅ I 2 n +
3

Example 3:
The velocity v(t) of a rocket from t1=8 s to t2=30 is given by:
⎡ 140000 ⎤
v(t ) = 2000 ln ⎢ ⎥ − 9.8t
⎣140000 − 2100t ⎦
a) Use Richardson extrapolation rule to find the distance covered

for n = 2
b) Find the relative true error

Table of results for n = 8 segments (trapezoidal rule)
n ∆x Et ∈t % ∈a %
1 11868 -807 7.296 ---
2 11266 -205 1.853 5.343
3 11153 -91.4 0.8265 1.019
4 11113 -51.5 0.4655 0.3594
5 11094 -33.0 0.2981 0.1669
6 11084 -22.9 0.2070 0.09082
7 11078 -16.8 0.1521 0.05482
8 11074 -12.9 0.1165 0.03560

a)
I 2 = 11266m I 4 = 11113m
I 2n − I n
TV ≅ I 2n + for n=2
3
I4 − I2 11113− 11266
TV ≅ I 4 + = 11113+
3 3
= 11062m

b) The true value:
⎛
30
⎡ 140000 ⎤ ⎞
x = ∫ ⎜ 2000 ln ⎢ ⎥ − 9 . 8t ⎟dt = 11061 m
8⎝ ⎣140000 − 2100t ⎦ ⎠
The absolute true error:
Et = 11061− 11062 = −1 m

c) The relative error:

11061 − 11062
∈t = × 100 = 0.00904%
11061
Comparison of different methods:
.
n ∆x (m) ∈t % ∆x (m) ∈t %
Trapezoidal Richardson
Trapezoidal rule Richardson
rule extrapolation
extrapolation
1 11868 7.296 -- --
2 11266 1.854 11065 0.03616
4 11113 0.4655 11062 0.009041
8 11074 0.1165 11061 0.0000

Romberg's method
Romberg’s method uses the same pattern as Richardson

extrapolation. However, Romberg used a recursive algorithm for
the extrapolation as follows:
I 2n − I n
TV ≅ I 2 n +
3
The true value TV is replaced by the result of the
(I 2n )R
Note also that the sign ≅ is replaced by the sign =
I 2n − I n I 2n − I n
(I 2n )R = I 2n + = I 2n + 2−1
3 4 −1

Romberg's method
Esimated true value is given by: TV ≅ ( I 2 n ) R + Ch 4

where: Ch4 is the value of the error of approximation
Another value of integral obtained while doubling the number of
segments from 2n to 4n:
I 4n − I 2n
(I 4n )R = I 4n +
3
Estimated true value is given by:
(I 4n )R − (I 2n )R
TV ≅ (I 4 n )R +
15
(I 4 n )R − (I 2 n )R
= ( I 4n ) R +
43−1 − 1
Romberg's method
A general expression for Romberg integration can be written as:
I k −1, j +1 − I k −1, j
I k , j = I k −1, j +1 + k −1
,k ≥ 2
4 −1
The index k represents the order of extrapolation
k=1 represents the values obtained from the regular trapezoidal rule
k=2 represents the values obtained using the true error estimate as O(h2)
The value of an integral with for j+ 1 is more accurate than

the value of the integral for j index

Romberg's method
Example 4:
⎡ 140000 ⎤
v(t ) = 2000 ln ⎢ ⎥ − 9.8t
⎣140000 − 2100t ⎦
a) Use Romberg’s method to find the distance covered. Use

n = 1, 2, 4, and 8
b) Find the absolute true error and the relative approximate error

Table of results for n = 8 segments (trapezoidal rule)
n ∆x Et ∈t % ∈a %
1 11868 -807 7.296 ---
2 11266 -205 1.853 5.343
3 11153 -91.4 0.8265 1.019
4 11113 -51.5 0.4655 0.3594
5 11094 -33.0 0.2981 0.1669
6 11084 -22.9 0.2070 0.09082
7 11078 -16.8 0.1521 0.05482
8 11074 -12.9 0.1165 0.03560

Romberg's method
From this table, the initial values from the trapezoidal rule are:
I 1,1 = 11868 I 1,2 = 11266
I 1,3 = 11113 I 1,4 = 11074
To get the first order extrapolation values:
I1,2 − I1,1
I 2 ,1 = I1,2 +
3
11266 − 11868
= 11266 +
3

Romberg's method
Similarly,
I 1,3 − I 1, 2 I 1, 4 − I 1,3
I 2, 2 = I 1,3 + I 2,3 = I 1, 4 +
3 3
11113 − 11266 11074 − 11113

= 11113 + = 11074 +
3 3
= 11062 = 11061

Romberg's method
For the second order extrapolation:
I 2 ,2 − I 2 ,1
I 3 ,1 = I 2 ,2 +
15
11062 − 11065
= 11062 +
15
= 11062
I 2,3 − I 2, 2
Similarly, I 3, 2 = I 2,3 +
15
11061 − 11062
= 11061 +
15
= 11061
Romberg's method
For the third order extrapolation
I 3 ,2 − I 3,1
I 4 ,1 = I 3 ,2 +
63
11061 − 11062
= 11061 +
63
= 11061m

Romberg's method
Appr. 1 Appr. 2 Appr. 3

1-segment 11868
11065
2-segment 11262 11062
11062 11061
4-segment 11113 11061
11061
8-segment 11074
Improved estimates of the value of an integral using Romberg

integration

Simpson's rule
The trapezoidal rule was based on approximating the integrand

by a first order polynomial, and then integrating the polynomial
over an interval from a to b. Simpson’s rule assumes that the
integrand can be approximated by a second order polynomial.
b b
I = ∫ f ( x )dx ≈ ∫ f 2 ( x )dx
a a
where:
f 2 ( x ) = a0 + a1 x + a2 x 2

Simpson's rule
A parabola passing through Y

three points : f(x)
( a , f ( a )),
⎛ a + b ⎛ a + b ⎞⎞
⎜ ,f⎜ ⎟ ⎟, fI(x)
⎝ 2 ⎝ 2 ⎠⎠
( b , f ( b ))
f ( a ) = f 2 ( a ) = a 0 + a1 a + a 2 a 2
a b X
2
⎛a + b⎞ ⎛a + b⎞ ⎛a + b⎞ ⎛a + b⎞
f⎜ ⎟ = f2 ⎜ ⎟ = a0 + a1 ⎜ ⎟ + a2 ⎜ ⎟
⎝ 2 ⎠ ⎝ 2 ⎠ ⎝ 2 ⎠ ⎝ 2 ⎠
f ( b ) = f 2 ( b ) = a 0 + a1b + a 2 b 2

Simpson's rule
Coefficients a0,a1,a2 are:
2 ⎛a + b⎞ 2
a f ( b ) + abf ( b ) − 4abf ⎜ ⎟ + abf ( a ) + b f(a)
a0 = ⎝ 2 ⎠
a 2 − 2ab + b 2
⎛a + b⎞ ⎛a + b⎞
af ( a ) − 4af ⎜ ⎟ + 3af ( b ) + 3bf ( a ) − 4bf ⎜ ⎟ + bf ( b )
a1 = − ⎝ 2 ⎠ ⎝ 2 ⎠
a 2 − 2ab + b 2
⎛ ⎛a + b⎞ ⎞
2⎜ f ( a ) − 2 f ⎜ ⎟ + f ( b ) ⎟
⎝ ⎝ 2 ⎠ ⎠
a2 =
a 2 − 2ab + b 2

Simpson's rule
Since: b
I ≈ ∫ f 2 ( x )dx
a
= ∫ (a0 + a1 x + a2 x 2 )dx
b
3 b
⎡ x x ⎤
2
= ⎢a0 x + a1 + a2 ⎥
⎣ 2 3 ⎦a
b2 − a2 b3 − a 3
= a0 ( b − a ) + a1 + a2
2 3

Simpson's rule
b
b−a⎡ ⎛a + b⎞ ⎤
∫ f 2 ( x )dx = ⎢ f(a)+ 4 f ⎜ ⎟ + f ( b )⎥
a 6 ⎣ ⎝ 2 ⎠ ⎦
b−a
h=
2
It follows that:
h⎡ ⎛ a+b⎞ ⎤
b
∫
a
f 2 ( x)dx = ⎢ f (a ) + 4 f ⎜
3⎣ ⎝ 2 ⎠
⎟ + f (b)⎥
⎦
it is called Simpson’s 1/3 rule

Multi-segment Simpson's rule
b x2 x4
f(x)
∫ f ( x )dx = ∫ f ( x )dx + ∫ f ( x )dx + .....
a x0 x2
xn − 2 xn
.... + ∫ f ( x )dx + ∫ f ( x )dx . . .
xn − 4 xn − 2
x
b
⎡ f ( x0 ) + 4 f ( x1 ) + f ( x2 )⎤ x0 x2 xn-2 xn
∫ f ( x )dx = ( x2 − x0 )⎢ ⎥ + ...
a ⎣ 6 ⎦ xi − xi − 2 = 2 h
⎡ f ( x2 ) + 4 f ( x3 ) + f ( x4 )⎤ i = 2, 4, ..., n
+ ( x4 − x2 )⎢ ⎥ + ...
⎣ 6 ⎦
⎡ f ( xn − 4 ) + 4 f ( xn − 3 ) + f ( xn − 2 ) ⎤
... + ( xn − 2 − xn − 4 )⎢ ⎥ + ...
⎣ 6 ⎦
⎡ f ( xn − 2 ) + 4 f ( xn −1 ) + f ( xn )⎤
+ ( xn − xn − 2 )⎢ ⎥⎦
⎣ 6
Simpson's rule
b
⎡ f ( x0 ) + 4 f ( x1 ) + f ( x2 )⎤
∫ f ( x )dx = 2h ⎢ ⎥ + ...
a ⎣ 6 ⎦
⎡ f ( x2 ) + 4 f ( x3 ) + f ( x4 )⎤
+ 2h ⎢ ⎥ + ...
⎣ 6 ⎦
⎡ f ( xn − 4 ) + 4 f ( xn − 3 ) + f ( x n − 2 ) ⎤
+ 2h ⎢ ⎥ + ...
⎣ 6 ⎦
⎡ f ( xn − 2 ) + 4 f ( xn −1 ) + f ( xn )⎤
+ 2h ⎢ ⎥⎦
⎣ 6

Simpson's rule
b h
∫ f ( x )dx = 3 [ f ( x0 ) + 4{ f ( x1 ) + f ( x3 ) + ... + f ( xn −1 )} + ...]
a
... + 2{ f ( x2 ) + f ( x4 ) + ... + f ( xn − 2 )} + f ( xn )}]

⎡ −1 − 2
⎤
h⎢ n n
= f ( x 0 ) + 4 ∑ f ( xi ) + 2 ∑ f ( xi ) + f ( x n ) ⎥
3⎢ i =1 i =2 ⎥
⎣ i = odd i = even ⎦
⎡ ⎤
b−a⎢ n −1 n − 2
= f ( x 0 ) + 4 ∑ f ( xi ) + 2 ∑ f ( xi ) + f ( x n ) ⎥
3n ⎢ i =1 i =2 ⎥
⎣ i = odd i = even ⎦

Estimation of errors in Simpson's rule
Approximate values of the integral,

using Simpson's rule with multiple segments
n Approximate values Et |Єt |
2 11065.72 4.38 0.0396%

4 11061.64 0.30 0.0027%
6 11061.40 0.06 0.0005%
8 11061.35 0.01 0.0001%
10 11061.34 0.00 0.0000%

Simpson's rule errors
(b − a) 5 ( 4)
Error for one segment Et = − f (ζ ), a < ζ < b
2880
Error for the multi-segment
( x2 − x0 ) ( 4 )
5 h5 ( 4 ) x0 < ζ1 < x2
E1 = − f ( ζ1 ) = − f ( ζ1 ),
2880 90
( x4 − x2 )5 ( 4 ) h5 ( 4 )
E2 = − f ( ζ2 ) = − f ( ζ 2 ), x2 < ζ 2 < x4
2880 90
( x2i − x2( i −1 ) )5 h 5
Ei = − f ( 4 ) ( ζ i ) = − f ( 4 ) ( ζ i ), x2( i −1 ) < ζ i < x2i
2880 90

Simpson's rule errors
True error
n n n
5 5
=
2
−
h
∑
2
( 4)
( ζ ) (b − a ) 2
Et = ∑ Ei f i =− ∑ f ( 4)
(ζ i )
i =1 90 i =1 5
90n i =1
n
2 ( 4)
∑f (ζ i )
(b − a ) 5
i =1
=−
90n 4 n
n
2
∑f
( 4)
(ζ i ) (b − a) 5 ( 4)
( 4) Et = − f
f = i =1 90n 4
n
the average value of the derivative

Gauss-Quadrature Method
Gaussian integral is given by:
b
I = ∫ f ( x )dx ≈ c1 f ( x1 ) + c2 f ( x2 )
a
constant coefficients
Points x1 and x2, which define the value of the integrand are not fixed
(as before), but there are a priori distributed randomly within <a,b>.

There are four unknowns x1, x2, c1, c2. These are found by
assuming that the formula gives exact results for integrating a
general third order polynomial:
f ( x ) = a0 + a1 x + a2 x 2 + a3 x 3 .
( )dx
b b
2 3
∫ f ( x )dx = a +
∫ 0 1 a x + a 2 x + a 3 x
a a b
⎡ x 2
x x ⎤ 3 4
= ⎢a0 x + a1 + a 2 + a3 ⎥
⎣ 2 3 4 ⎦a
⎛ b2 − a2 ⎞ ⎛ b3 − a3 ⎞ ⎛ b4 − a4 ⎞
= a0 (b − a ) + a1 ⎜⎜ ⎟⎟ + a 2 ⎜⎜ ⎟⎟ + a3 ⎜⎜ ⎟⎟
⎝ 2 ⎠ ⎝ 3 ⎠ ⎝ 4 ⎠

b
I = ∫ f ( x )dx ≈ c1 f ( x1 ) + c2 f ( x2 )
a
Hence:
∫ f ( x )dx = c1 (a0 + a1 x1 + a 2 x1 + a3 x1 ) + c2 (a0 + a1 x2 + a 2 x2 + a3 x2 )

b
2 3 2 3
( ) (
= a0 (c1 + c2 ) + a1 (c1 x1 + c2 x2 ) + a 2 c1 x1 + c2 x2 + a3 c1 x1 + c2 x2
2 2 3 3
)
The formula would then give:
( )dx
b b
2 3
∫ f ( x )dx = ∫ 0 1
a + a x + a 2 x + a 3 x
a a
⎛ b2 − a2 ⎞ ⎛ b3 − a3 ⎞ ⎛ b4 − a4 ⎞
= a0 (b − a ) + a1 ⎜⎜ ⎟⎟ + a 2 ⎜⎜ ⎟⎟ + a3 ⎜⎜ ⎟⎟
⎝ 2 ⎠ ⎝ 3 ⎠ ⎝ 4 ⎠

( ) (
a0 (c1 + c2 ) + a1 (c1 x1 + c2 x2 ) + a2 c1 x12 + c2 x2 2 + a3 c1 x13 + c2 x2 3 )
⎛ b2 − a2 ⎞ ⎛ b3 − a3 ⎞ ⎛ b4 − a4 ⎞
= a0 (b − a ) + a1 ⎜⎜ ⎟⎟ + a 2 ⎜⎜ ⎟⎟ + a3 ⎜⎜ ⎟⎟
⎝ 2 ⎠ ⎝ 3 ⎠ ⎝ 4 ⎠
b − a = c1 + c2 b2 − a2
= c1 x1 + c 2 x 2
2
b3 − a3 2 2 b4 − a4
= c1 x1 + c2 x2 3
= c1 x1 + c 2 x 2
3
3 4

This gives us four equations as follows:

b − a = c1 + c2
b2 − a2
= c1 x1 + c 2 x 2
2
b − a3
3
2 2
= c1 x1 + c2 x2
3
b4 − a4 3 3
= c1 x1 + c 2 x 2
4
we can find that the above four simultaneous nonlinear equations
have only one acceptable solution
⎛ b − a ⎞⎛ 1 ⎞ b + a ⎛ b − a ⎞⎛ 1 ⎞ b + a
x1 = ⎜ ⎟⎜ − ⎟+ x2 = ⎜ ⎟⎜ ⎟ +
⎝ 2 ⎠⎝ 3⎠ 2 ⎝ 2 ⎠⎝ 3 ⎠ 2
b−a b−a
c1 = c2 =
2 2
Hence:
∫ f ( x )dx ≈ c1 f ( x1 ) + c2 f ( x2 )
a
b−a ⎛b−a ⎛ 1 ⎞ b+a⎞ b−a ⎛b−a ⎛ 1 ⎞ b+a⎞

= f ⎜⎜ ⎜− ⎟+ ⎟⎟ + f ⎜⎜ ⎜ ⎟+ ⎟⎟
2 ⎝ 2 ⎝ 3⎠ 2 ⎠ 2 ⎝ 2 ⎝ 3⎠ 2 ⎠
General n-point rules would approximate the integral:
b
∫ f ( x )dx ≈ c1 f ( x1 ) + c 2 f ( x 2 ) + . . . . . . . + c n f ( x n )
a

The coefficients and arguments for n-point Gauss method are

given within the range of <-1,1> :
1 n
∫ g ( x )dx ≅ ∑ ci g ( xi )
−1 i =1
n Coefficients Function arguments
2 c1 = 1.000000000 x1 = -0.577350269
c2 = 1.000000000 x2 = 0.577350269
3 c1 = 0.555555556 x1 = -0.774596669
c2 = 0.888888889 x2 = 0.000000000
c3 = 0.555555556 x3 = 0.774596669
4 c1 = 0.347854845 x1 = -0.861136312
c2 = 0.652145155 x2 = -0.339981044
c3 = 0.652145155 x3 = 0.339981044
c4 = 0.347854845 x4 = 0.861136312

n Coefficients Function arguments
5 c1 = 0.236926885 x1 = -0.906179846
c2 = 0.478628670 x2 = -0.538469310
c3 = 0.568888889 x3 = 0.000000000
c4 = 0.478628670 x4 = 0.538469310
c5 = 0.236926885 x5 = 0.906179846
6 c1 = 0.171324492 x1 = -0.932469514
c2 = 0.360761573 x2 = -0.661209386
c3 = 0.467913935 x3 = -0.2386191860
c4 = 0.467913935 x4 = 0.2386191860
c5 = 0.360761573 x5 = 0.661209386
c6 = 0.171324492 x6 = 0.932469514

1
So if the table is given for: ∫ g ( x )dx integrals, how does one solve
−1
b
∫ f ( x )dx ?
[a , b]
a
The answer lies in that any integral with limits of
can be converted into an integral with limits: [− 1, 1]
Let, x = mt + c
It follows that:
if x = a, t = −1
b−a
m=
ir x = b, t= 1 2
b+a
c=
2
b−a b+a
Hence: x= t+
2 2
b−a
dx = dt
2
Substituting our values of x and dx into the integral gives us:
1
⎛b−a b+a⎞b−a
b
∫
a
f ( x )dx = ∫ f ⎜
−1 ⎝
2
t+ ⎟
2 ⎠ 2
dt

Example 5:
Use two-point Gauss quadrature rule to approximate the distance
covered by a rocket from t = 8 s to t = 30 s if the velocity is given
by:
⎡ 140000 ⎤
v(t ) = 2000 ln ⎢ ⎥ − 9.8t
⎣140000 − 2100t ⎦
a) Use the Gauss-Quadrature Mmethod to find the distance

covered from t1=8 s to t2=30 s
b) Find the true error.

First, change the limits of integration from from [8,30] to [-1,1]
30
30 − 8 1 ⎛ 30 − 8 30 + 8 ⎞
∫ f ( t )dt = ∫ f⎜ x+ ⎟dx
8 2 −1 ⎝ 2 2 ⎠
1
= 11 ∫ f (11x + 19 )dx
−1
The weighting factors and function c1 = 1 .000000000
argument values are (n=2):
x1 = − 0 .577350269
c 2 = 1 .000000000
x 2 = 0 .577350269

The formula is:
1
11 ∫ f (11x + 19 )dx ≈ 11c1 f (11x1 + 19 ) + 11c 2 f (11x 2 + 19 )
−1
= 11 f (11( −0.5773503 ) + 19) + 11 f (11( 0.5773503 ) + 19)
= 11 f ( 12.64915 ) + 11 f ( 25.35085 )
= 11( 296.8317 ) + 11( 708.4811 )
= 11058.44 m

Since:
⎡ 140000 ⎤
f ( 12.64915 ) = 2000 ln ⎢ ⎥ − 9.8( 12.64915 )
⎣140000 − 2100( 12.64915 )⎦
= 296.8317
⎡ 140000 ⎤
f ( 25.35085 ) = 2000 ln ⎢ ⎥ − 9.8( 25.35085 )
⎣140000 − 2100( 25.35085 )⎦
= 708.4811

b) The absolute true error:
Et = 11061.34 − 11058.44
= 2.9000 m
c) The relative error: ∈t
11061.34 − 11058.44
∈t = × 100%
11061.34
= 0.0262%

Integracion Richardson y Roomberg

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Integracion Richardson y Roomberg

Uploaded by

Copyright:

Available Formats

Numerical Methods

dr hab. inż. Katarzyna Zakrzewska, prof. AGH

Numerical Methods lecture 5 1

Numerical Methods lecture 5 2

The integral can

Numerical Methods lecture 5 3

Newton – Cotes integration belongs to a class of

Then, the integral of f(x) can be approximated as an

Numerical Methods lecture 5 4

The trapezoidal rule f1 ( x) = a0 + a1 x

Numerical Methods lecture 5 6

a) Use the single segment trapezoidal rule to find the distance

Numerical Methods lecture 5 7

⎡177.27 + 901.67 ⎤ = 11868 m

Numerical Methods lecture 5 8

b) The true value

The relative error:

Numerical Methods lecture 5 9

The true error using a

b a+h a+2h a + 3h a+4h

Numerical Methods lecture 5 10

Numerical Methods lecture 5 11

a) Use the complex segment trapezoidal rule to find the distance

Numerical Methods lecture 5 12

Numerical Methods lecture 5 13

The relative error:

Numerical Methods lecture 5 14

Numerical Methods lecture 5 15

The relative error for a simple trapezoidal rule is

Numerical Methods lecture 5 16

Numerical Methods lecture 5 17

Table below presents the results for the integral

4 11113 -51.5 0.4655 0.3594

8 11074 -12.9 0.1165 0.03560

16 11065 -3.22 0.02913 0.00401

Numerical Methods lecture 5 19

Richardson’s extrapolation and Romberg’s method

Numerical Methods lecture 5 20

The true error obtained when using the multi-segment

approximate value using

Numerical Methods lecture 5 21

If the number of segments is doubled from n to 2n:

Numerical Methods lecture 5 22

The velocity v(t) of a rocket from t1=8 s to t2=30 is given by:

a) Use Richardson extrapolation rule to find the distance covered

Numerical Methods lecture 5 23

Numerical Methods lecture 5 24

Numerical Methods lecture 5 25

b) The true value:

The absolute true error:

Numerical Methods lecture 5 26

c) The relative error:

Numerical Methods lecture 5 27

Romberg’s method uses the same pattern as Richardson

Numerical Methods lecture 5 28

Esimated true value is given by: TV ≅ ( I 2 n ) R + Ch 4

A general expression for Romberg integration can be written as:

The index k represents the order of extrapolation

The value of an integral with for j+ 1 is more accurate than

Numerical Methods lecture 5 30

Velocity v(t) of a rocket from t1=8 s to t2=30 s is given by:

a) Use Romberg’s method to find the distance covered. Use

Numerical Methods lecture 5 31

Numerical Methods lecture 5 32